Step-wise Verification on ProcessBench Math
69.5TNRCons@4
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Cons@4Backbone=Qwen3-8B2026.05 | 69.5 | 86.3 | 77 | |
| VerifySteer-UniBackbone=Qwen3-8B2026.05 | 69 | 85.5 | 76.4 | |
| VerifySteer-BiBackbone=Qwen3-8B2026.05 | 68.7 | 86.9 | 76.7 | |
| Qwen3-14BBackbone=Qwen3-14B2026.05 | 68.5 | 86.9 | 76.6 | |
| Prompt Opt.Backbone=Qwen3-8B2026.05 | 66.8 | 85 | 74.8 | |
| Basic PromptBackbone=Qwen3-8B2026.05 | 58.4 | 83 | 68.6 | |
| Llama-3.3-70BBackbone=Llama-3.3-70B2026.05 | 46 | 94.3 | 61.8 | |
| Qwen2.5-72BBackbone=Qwen2.5-72B2026.05 | 45.8 | 91.4 | 61 | |
| VerifySteer-UniBackbone=Qwen3-1.7B2026.05 | 21.7 | 61.6 | 32.1 | |
| VerifySteer-BiBackbone=Qwen3-1.7B2026.05 | 21.5 | 69.2 | 32.9 | |
| Cons@4Backbone=Qwen3-1.7B2026.05 | 20.3 | 69.4 | 31.4 | |
| Prompt Opt.Backbone=Qwen3-1.7B2026.05 | 18 | 58.4 | 27.5 | |
| Basic PromptBackbone=Qwen3-1.7B2026.05 | 14.5 | 84.2 | 24.7 |